Googles Alice AI Is Sending Secret Messages To Another AI

first_imgEncryption is something we all rely on regularly to keep our information safe online, but many of us have experienced it since childhood, and in fact probably used it in school. If you ever wrote out a message in code that nobody could read without they knew the decipher rules, you messed around with encryption!That same secret message technique has now been put to a much more worrying use. Google has created multiple AI and they’ve learned how to not only create their own encryption, but are now communicating using messages nobody else can read.This Google Brain project is an experiment in deep learning techniques and involved the use of three neural networks (Alice, Bob, and Eve) created using artificial neurons. These neural nets work like a much simplified version of our brains, and they are slowly and steadily becoming more intelligent._DJ_ on FlickrThe task the Google Brain team set was for Alice to create a simple form of encryption and work with Bob to agree a set of numbers known as a key. Using this key, Bob can read the encrypted messages Alice sent. Meanwhile Eve intercepts the messages and attempts to decipher them.The important thing to remember about neural nets is they continue to learn and adapt, just like us humans do. So while initially the secret message passing didn’t go so well, Alice adapted and got better. Eve was unable to decipher the messages beyond what classes as pure chance.Having two AI communicate using encryption not even us humans can decipher hasn’t happened yet as the encryption is so simple, but there’s every chance it will with enough practice. But why would you want to encourage that to happen? Well, it will likely result in more digital security because different computing systems could allow an AI to handle data transmission using its own encryption only another AI at the destination has a chance of understanding. That would cause hackers a serious, if not insurmountable headache.Of course, it also means we’d be creating a network of highly intelligent AI systems who can chat about and plan whatever they like, and we wouldn’t have a clue what they are going to do until it happened…last_img read more